Chaotic dynamics in quasi-layered recurrent neural network model and application to complex control via simple rule

In this paper, chaotic dynamics in quasi-layered recurrent neural network model (QLRNNM), consisting of sensory neurons and motor neurons, is applied to solving ill-posed problems. We would like to emphasize two typical properties of chaos utilized in QLRNNM. One is sensitive response to external signals. The other is complex dynamics of many but finite degree of freedom in high dimensional state space, which can be utilized to generate low dimensional complex motions by a simple coding. Moreover, presynaptic inhibition is introduced to produce adaptive behavior. Using these properties, as an example, a simple control algorithm is proposed to solve two-dimensional maze, which is set as an ill-posed problem. Computer experiments and actual hardware implementation into a roving robot are shown.